1       SUBROUTINE ZUNGQL( M, N, K, A, LDA, TAU, WORK, LWORK, INFO )
  2 *
  3 *  -- LAPACK routine (version 3.2) --
  4 *  -- LAPACK is a software package provided by Univ. of Tennessee,    --
  5 *  -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
  6 *     November 2006
  7 *
  8 *     .. Scalar Arguments ..
  9       INTEGER            INFO, K, LDA, LWORK, M, N
 10 *     ..
 11 *     .. Array Arguments ..
 12       COMPLEX*16         A( LDA, * ), TAU( * ), WORK( * )
 13 *     ..
 14 *
 15 *  Purpose
 16 *  =======
 17 *
 18 *  ZUNGQL generates an M-by-N complex matrix Q with orthonormal columns,
 19 *  which is defined as the last N columns of a product of K elementary
 20 *  reflectors of order M
 21 *
 22 *        Q  =  H(k) . . . H(2) H(1)
 23 *
 24 *  as returned by ZGEQLF.
 25 *
 26 *  Arguments
 27 *  =========
 28 *
 29 *  M       (input) INTEGER
 30 *          The number of rows of the matrix Q. M >= 0.
 31 *
 32 *  N       (input) INTEGER
 33 *          The number of columns of the matrix Q. M >= N >= 0.
 34 *
 35 *  K       (input) INTEGER
 36 *          The number of elementary reflectors whose product defines the
 37 *          matrix Q. N >= K >= 0.
 38 *
 39 *  A       (input/output) COMPLEX*16 array, dimension (LDA,N)
 40 *          On entry, the (n-k+i)-th column must contain the vector which
 41 *          defines the elementary reflector H(i), for i = 1,2,...,k, as
 42 *          returned by ZGEQLF in the last k columns of its array
 43 *          argument A.
 44 *          On exit, the M-by-N matrix Q.
 45 *
 46 *  LDA     (input) INTEGER
 47 *          The first dimension of the array A. LDA >= max(1,M).
 48 *
 49 *  TAU     (input) COMPLEX*16 array, dimension (K)
 50 *          TAU(i) must contain the scalar factor of the elementary
 51 *          reflector H(i), as returned by ZGEQLF.
 52 *
 53 *  WORK    (workspace/output) COMPLEX*16 array, dimension (MAX(1,LWORK))
 54 *          On exit, if INFO = 0, WORK(1) returns the optimal LWORK.
 55 *
 56 *  LWORK   (input) INTEGER
 57 *          The dimension of the array WORK. LWORK >= max(1,N).
 58 *          For optimum performance LWORK >= N*NB, where NB is the
 59 *          optimal blocksize.
 60 *
 61 *          If LWORK = -1, then a workspace query is assumed; the routine
 62 *          only calculates the optimal size of the WORK array, returns
 63 *          this value as the first entry of the WORK array, and no error
 64 *          message related to LWORK is issued by XERBLA.
 65 *
 66 *  INFO    (output) INTEGER
 67 *          = 0:  successful exit
 68 *          < 0:  if INFO = -i, the i-th argument has an illegal value
 69 *
 70 *  =====================================================================
 71 *
 72 *     .. Parameters ..
 73       COMPLEX*16         ZERO
 74       PARAMETER          ( ZERO = ( 0.0D+00.0D+0 ) )
 75 *     ..
 76 *     .. Local Scalars ..
 77       LOGICAL            LQUERY
 78       INTEGER            I, IB, IINFO, IWS, J, KK, L, LDWORK, LWKOPT,
 79      $                   NB, NBMIN, NX
 80 *     ..
 81 *     .. External Subroutines ..
 82       EXTERNAL           XERBLA, ZLARFB, ZLARFT, ZUNG2L
 83 *     ..
 84 *     .. Intrinsic Functions ..
 85       INTRINSIC          MAXMIN
 86 *     ..
 87 *     .. External Functions ..
 88       INTEGER            ILAENV
 89       EXTERNAL           ILAENV
 90 *     ..
 91 *     .. Executable Statements ..
 92 *
 93 *     Test the input arguments
 94 *
 95       INFO = 0
 96       LQUERY = ( LWORK.EQ.-1 )
 97       IF( M.LT.0 ) THEN
 98          INFO = -1
 99       ELSE IF( N.LT.0 .OR. N.GT.M ) THEN
100          INFO = -2
101       ELSE IF( K.LT.0 .OR. K.GT.N ) THEN
102          INFO = -3
103       ELSE IF( LDA.LT.MAX1, M ) ) THEN
104          INFO = -5
105       END IF
106 *
107       IF( INFO.EQ.0 ) THEN
108          IF( N.EQ.0 ) THEN
109             LWKOPT = 1
110          ELSE
111             NB = ILAENV( 1'ZUNGQL'' ', M, N, K, -1 )
112             LWKOPT = N*NB
113          END IF
114          WORK( 1 ) = LWKOPT
115 *
116          IF( LWORK.LT.MAX1, N ) .AND. .NOT.LQUERY ) THEN
117             INFO = -8
118          END IF
119       END IF
120 *
121       IF( INFO.NE.0 ) THEN
122          CALL XERBLA( 'ZUNGQL'-INFO )
123          RETURN
124       ELSE IF( LQUERY ) THEN
125          RETURN
126       END IF
127 *
128 *     Quick return if possible
129 *
130       IF( N.LE.0 ) THEN
131          RETURN
132       END IF
133 *
134       NBMIN = 2
135       NX = 0
136       IWS = N
137       IF( NB.GT.1 .AND. NB.LT.K ) THEN
138 *
139 *        Determine when to cross over from blocked to unblocked code.
140 *
141          NX = MAX0, ILAENV( 3'ZUNGQL'' ', M, N, K, -1 ) )
142          IF( NX.LT.K ) THEN
143 *
144 *           Determine if workspace is large enough for blocked code.
145 *
146             LDWORK = N
147             IWS = LDWORK*NB
148             IF( LWORK.LT.IWS ) THEN
149 *
150 *              Not enough workspace to use optimal NB:  reduce NB and
151 *              determine the minimum value of NB.
152 *
153                NB = LWORK / LDWORK
154                NBMIN = MAX2, ILAENV( 2'ZUNGQL'' ', M, N, K, -1 ) )
155             END IF
156          END IF
157       END IF
158 *
159       IF( NB.GE.NBMIN .AND. NB.LT..AND. NX.LT.K ) THEN
160 *
161 *        Use blocked code after the first block.
162 *        The last kk columns are handled by the block method.
163 *
164          KK = MIN( K, ( ( K-NX+NB-1 ) / NB )*NB )
165 *
166 *        Set A(m-kk+1:m,1:n-kk) to zero.
167 *
168          DO 20 J = 1, N - KK
169             DO 10 I = M - KK + 1, M
170                A( I, J ) = ZERO
171    10       CONTINUE
172    20    CONTINUE
173       ELSE
174          KK = 0
175       END IF
176 *
177 *     Use unblocked code for the first or only block.
178 *
179       CALL ZUNG2L( M-KK, N-KK, K-KK, A, LDA, TAU, WORK, IINFO )
180 *
181       IF( KK.GT.0 ) THEN
182 *
183 *        Use blocked code
184 *
185          DO 50 I = K - KK + 1, K, NB
186             IB = MIN( NB, K-I+1 )
187             IF( N-K+I.GT.1 ) THEN
188 *
189 *              Form the triangular factor of the block reflector
190 *              H = H(i+ib-1) . . . H(i+1) H(i)
191 *
192                CALL ZLARFT( 'Backward''Columnwise', M-K+I+IB-1, IB,
193      $                      A( 1, N-K+I ), LDA, TAU( I ), WORK, LDWORK )
194 *
195 *              Apply H to A(1:m-k+i+ib-1,1:n-k+i-1) from the left
196 *
197                CALL ZLARFB( 'Left''No transpose''Backward',
198      $                      'Columnwise', M-K+I+IB-1, N-K+I-1, IB,
199      $                      A( 1, N-K+I ), LDA, WORK, LDWORK, A, LDA,
200      $                      WORK( IB+1 ), LDWORK )
201             END IF
202 *
203 *           Apply H to rows 1:m-k+i+ib-1 of current block
204 *
205             CALL ZUNG2L( M-K+I+IB-1, IB, IB, A( 1, N-K+I ), LDA,
206      $                   TAU( I ), WORK, IINFO )
207 *
208 *           Set rows m-k+i+ib:m of current block to zero
209 *
210             DO 40 J = N - K + I, N - K + I + IB - 1
211                DO 30 L = M - K + I + IB, M
212                   A( L, J ) = ZERO
213    30          CONTINUE
214    40       CONTINUE
215    50    CONTINUE
216       END IF
217 *
218       WORK( 1 ) = IWS
219       RETURN
220 *
221 *     End of ZUNGQL
222 *
223       END